I have similar project.
I did it with two voltages. First I put the switching regulator to drop to 3.9V for GSM. Then LDO to drop from 3.9V to 3.3V for MCU, GPS, CAN.
I see the GPS module you pointed out possibly has it's own LDO on board and that's why its voltage ranges from 3.3V. Take a look on those modules alone http://www.gtop-tech.com/en/category/GPS-Antenna-Module/A01_MT3339.html they all go from 3.0V so both voltages could be used (3.9V or 3.3V).
For CAN tranceiver I used SN65HVD230D from Texas Instruments, which is a little more expensive but works with 3.3V (that eliminated necessity of third 5V supply).
Don't use voltage dividers for supplying power. The load connected to this divider acts like another resistor in parallel with one from divider. If the load you connect is mcu or some module, the impedance is unknown and variable in time - you wont be able to achieve desired voltage.
Voltage dividers are good if you connect some high impedance line to it. Then almost no current flow through that line and the voltage at the divider stays at the level it was set. This should be fine in case you need to lower some logic levels (you mentioned 4050 which is a good solution).
You might also need some protection circuit at the power input, before the voltage regulator. Automotive power can have load dumps (100V on power line for about 400ms), positive and negative voltage transients. Cold crank condition can lower the voltage to 3V for 15ms. Jump start can raise it to 24V.
Batteries connected in series provide the sum of their voltage but at the current of the lowest rated battery.
Batteries in parallel (of the same voltage) provide the same voltage as one battery, but at the current capacity of all battery current capacities summed.
Your amplifier is rated at 20V/3A which means it draws 3A most likely at maximum volume. I would expect it draws much less at lower volumes. However, I will provide calculations based on the highest current draw.
"D" cells (batteries) store more energy (18000+ mAh) than "AA" cells (~2750 mAh), and therefore will last longer than "AA" cells for a given load.
If you connected 14 "D" cells in series, that would provide 21 volts (a 5% voltage increase should be tolerated by your amplifier, and as the batteries drain, they will be significantly lower than this voltage).
To address the current draw, you will need to connect additional sets of 14 "D" cells in parallel. Each set connected will increase the current capacity. With only one set, you have a theoretical maximum of 18 amp-hours, which means you could operate a 3A device for 6 hours. However, if you look at the datasheet for "D" cells, the mAh capacity is significantly reduced at high current draw. For example, at just 0.5 A, the mAh capacity is given as ~1500 mAh. This means that your 3A amplifier would only run for a half hour! It gets worse - because you are drawing much more than 0.5A, the life of the batteries is much less. In fact I would not expect the unit to work properly with only one set of batteries in series.
Ideally you would connect enough sets in parallel so that you are only drawing 25mA from each set, which gives the best-case mAh capacity. But this would require that you connect an impractical 120 sets; a total of 1680 "D" cells!
With 6 groups of 14 cells, each set would provide 500mA, and come much closer to providing the 1500 mAh capacity shown on the datasheet. I would expect a run time of about a half hour then.
Best Answer
There are definitely 3 terminal regulators with low dropout that suit this requirement.
See eg www.digikey.com and enter search terms.
As an example only The Seiko S816 works with an external transistor and not including decoupling capacitors requires TWO components total. Dropout is limited by the external transistor's characteristics. While a bipolar would usually be used, this would drive a suitable MOSFET and dropout voltages as sensibly low as desired would be possible.
Using Digikey's search also found
MIC29302 in stock $2.86/1
3A, 250 mV dropout typical at 1.5A. Agh/whoops - I now see I've arrived at the same device as Spehro :-). Search Digikey using their excellent parameter driven search. This and more are there.
Also
Semtech SC1592 in stock $1.82/1.
260 mV dropout at 3A BUT uses a special dual input supply mode - power conversion is low dropout but Vspply_control needs to be Vout + 1.5V. May or may not be useful.